Spelling suggestions: "subject:"highperfomance computing"" "subject:"highperformance computing""
1 |
Implementação do software MILC no estudo da QCD completa / Implementation of the MILC package in the study of full QCDFernando Henrique e Paula da Luz 12 March 2010 (has links)
A CromoDinâmica Quântica (QCD) é a teoria quântica de campos que descreve as interações fortes entre quarks, que são os constituintes fundamentais das partículas do núcleo atômico. Devido ao caráter peculiar destas interações, o estudo da QCD não pode ser realizado pelos métodos usuais em teorias quânticas de campos, baseados em expansões perturbativas. O estudo não-perturbativo da QCD a partir de primeiros princípios torna-se possível através da formulação de rede da teoria, que equivale a um modelo de mecânica estatística clássica, para o qual podem ser realizadas simulações numéricas através de métodos de Monte Carlo. A área de simulações numéricas da QCD representa uma das maiores aplicações atuais da computação de alto desempenho, sendo realizada nos principais centros computacionais do mundo. As grandes exigências do trabalho de pesquisa nesta área contribuíram inclusive para o desenvolvimento de novas arquiteturas computacionais. O uso de processamento paralelo é vital nessas simulações, principalmente nos casos em que está envolvida a simulação da chamada QCD completa, onde se consideram os efeitos dos quarks dinâmicos. Vários pacotes contendo implementações de algoritmos para o estudo da QCD começam a ser disponibilizados por grupos de pesquisa na área. Nosso foco neste trabalho é voltado para o pacote MILC. Além de fazer uma descrição detalhada da forma de utilização deste pacote, realizamos aqui um acompanhamento da evolução dos métodos empregados, desde o Método de Monte Carlo aplicado no algoritmo de Metropolis até a elaboração do algoritmo RHMC, introduzido recentemente. Fazemos uma comparação de e_ciência entre o RHMC e o algoritmo R, que foi o mais utilizado por décadas. / Quantum ChromoDinamics (QCD) is the quantum field theory that describes the strong interactions between quarks, which are the fundamental constituents of particles in the atomic nucleus. Due to the peculiar characteristic of these interactions, the study of QCD cannot be carried out by usual methods in quantum field theory, which are based on pertubative expansions. The non-pertubative study of QCD from first principles becomes possible through the lattice formulation of the theory, which is equivalent to a classical statistical mechanics model, which in turn can be carried out by numerical simulations using Monte Carlo methods. The field of numerical simulations of QCD is one of the main applications of high performance computing, and is perfomed in most major computational centers around the world. The demanding requirements needed in this field led also to the development of new computational architectures. The use of parallel processing is vital in these types of simulations, especially in cases that involve what is known as full QCD, where the effects of dynamic quarks are taken into account. Several packages with algorithms implemented for the study of QCD have been recently made available by research groups in this field. The focus of this work is the MILC package. Here we make a detailed description of how to use this package and a follow up of the used methods, from the Monte Carlo method applied in the Metropolis algorithm up to the development of the RHMC algorithm, recently introduced. Comparisons are made between the e_ciency of RHMC and the R algorithm, which was the most used in the past decades.
|
2 |
Implementação do software MILC no estudo da QCD completa / Implementation of the MILC package in the study of full QCDLuz, Fernando Henrique e Paula da 12 March 2010 (has links)
A CromoDinâmica Quântica (QCD) é a teoria quântica de campos que descreve as interações fortes entre quarks, que são os constituintes fundamentais das partículas do núcleo atômico. Devido ao caráter peculiar destas interações, o estudo da QCD não pode ser realizado pelos métodos usuais em teorias quânticas de campos, baseados em expansões perturbativas. O estudo não-perturbativo da QCD a partir de primeiros princípios torna-se possível através da formulação de rede da teoria, que equivale a um modelo de mecânica estatística clássica, para o qual podem ser realizadas simulações numéricas através de métodos de Monte Carlo. A área de simulações numéricas da QCD representa uma das maiores aplicações atuais da computação de alto desempenho, sendo realizada nos principais centros computacionais do mundo. As grandes exigências do trabalho de pesquisa nesta área contribuíram inclusive para o desenvolvimento de novas arquiteturas computacionais. O uso de processamento paralelo é vital nessas simulações, principalmente nos casos em que está envolvida a simulação da chamada QCD completa, onde se consideram os efeitos dos quarks dinâmicos. Vários pacotes contendo implementações de algoritmos para o estudo da QCD começam a ser disponibilizados por grupos de pesquisa na área. Nosso foco neste trabalho é voltado para o pacote MILC. Além de fazer uma descrição detalhada da forma de utilização deste pacote, realizamos aqui um acompanhamento da evolução dos métodos empregados, desde o Método de Monte Carlo aplicado no algoritmo de Metropolis até a elaboração do algoritmo RHMC, introduzido recentemente. Fazemos uma comparação de e_ciência entre o RHMC e o algoritmo R, que foi o mais utilizado por décadas. / Quantum ChromoDinamics (QCD) is the quantum field theory that describes the strong interactions between quarks, which are the fundamental constituents of particles in the atomic nucleus. Due to the peculiar characteristic of these interactions, the study of QCD cannot be carried out by usual methods in quantum field theory, which are based on pertubative expansions. The non-pertubative study of QCD from first principles becomes possible through the lattice formulation of the theory, which is equivalent to a classical statistical mechanics model, which in turn can be carried out by numerical simulations using Monte Carlo methods. The field of numerical simulations of QCD is one of the main applications of high performance computing, and is perfomed in most major computational centers around the world. The demanding requirements needed in this field led also to the development of new computational architectures. The use of parallel processing is vital in these types of simulations, especially in cases that involve what is known as full QCD, where the effects of dynamic quarks are taken into account. Several packages with algorithms implemented for the study of QCD have been recently made available by research groups in this field. The focus of this work is the MILC package. Here we make a detailed description of how to use this package and a follow up of the used methods, from the Monte Carlo method applied in the Metropolis algorithm up to the development of the RHMC algorithm, recently introduced. Comparisons are made between the e_ciency of RHMC and the R algorithm, which was the most used in the past decades.
|
3 |
Fluid-solid interaction in a non-convex granular media : application to rotating drums and packed bed reactors / Intéraction fluide-solide en milieux granulaires de particules non-convexes : application aux tambours tourants et réacteurs à lit fixeRakotonirina, Andriarimina 01 December 2016 (has links)
Cette thèse porte sur l'étude numérique des écoulements fluide-particules rencontrés dans l'industrie. Ces travaux se situent dans le cadre de la compréhension des phénomènes qui se déroulent dans des tambours tournants et réacteurs à lit fixe en présence de particules de forme non convexe. En effet, la forme des particules influence de manière importante la dynamique de ces milieux. A cet effet, nous nous sommes servis de la plateforme numérique parallèle Grans3D pour la dynamique des milieux granulaires et PeliGRIFF pour les écoulements multiphasiques. Dans la première partie de cette thèse, nous avons développé une nouvelle stratégie numérique qui permet de prendre en compte des particules de forme arbitrairement non convexe dans le solveur Grains3D. Elle consiste à décomposer une forme non convexe en plusieurs formes convexes quelconques. Nous avons nommé cette méthode « glued-convex ». Le modèle a été validé avec succès sur des résultats théoriques et expérimentaux de tambours tournants en présence de particules en forme de croix. Nous avons aussi utilisé le modèle pour simuler le chargement de réacteurs à lits fixes puis des lois de corrélation sur les taux de vide ont été déduites de nos résultats numériques. Dans ces travaux, nous avons aussi testé les performances parallèles de nos outils sur des simulations numériques à grande échelle de divers systèmes de particules convexes. La deuxième partie de cette thèse a été consacrée à l'extension du solveur PeliGRIFF à pouvoir prendre en compte la présence de particules multilobées (non convexes) dans des écoulements monophasiques. Une approche du type Simulation Numérique Directe, basée sur les Multiplicateurs de Lagrange Distribués / Domaine Fictif (DLM/FD), a alors été adoptée pour résoudre l'écoulement autour des particules. Une série d'études de convergence spatiale a été faite basée sur diverses configurations et divers régimes. Enfin, ces outils ont été utilisés pour simuler des écoulements au travers de lits fixes de particules de forme multi-lobée dans le but d'étudier l'influence de la forme des particules sur l'hydrodynamique dans ces lits. Les résultats ont montré une consistance avec les résultats expérimentaux disponibles dans la littérature. / Non convex granular media are involved in many industrial processes as, e.g., particle calcination/drying in rotating drums or solid catalyst particles in chemical reactors. In the case of optimizing the shape of catalysts, the experimental discrimination of new shapes based on packing density and pressure drop proved to be difficult due to the limited control of size distribution and loading procedure. There is therefore a strong interest in developing numerical tools to predict the dynamics of granular media made of particles of arbitrary shape and to simulate the flow of a fluid (either liquid or gas) around these particles. Non-convex particles are even more challenging than convex particles due to the potential multiplicity of contact points between two solid bodies. In this work, we implement new numerical strategies in our home made high-fidelity parallel numerical tools: Grains3D for granular dynamics of solid particles and PeliGRIFF for reactive fluid/solid flows. The first part of this work consists in extending the modelling capabilities of Grains3D from convex to non-convex particles based on the decomposition of a non-convex shape into a set of convex particles. We validate our numerical model with existing analytical solutions and experimental data on a rotating drum filled with 2D cross particle shapes. We also use Grains3D to study the loading of semi-periodic small size reactors with trilobic and quadralobic particles. The second part of this work consists in extending the modelling capabilities of PeliGRIFF to handle poly-lobed (and hence non-convex) particles. Our Particle Resolved Simulation (PRS) method is based on a Distributed Lagrange Multiplier / Fictitious Domain (DLM/FD) formulation combined with a Finite Volume / Staggered Grid (FV/SG) discretization scheme. Due to the lack of analytical solutions and experimental data, we assess the accuracy of our PRS method by examining the space convergence of the computed solution in assorted flow configurations such as the flow through a periodic array of poly-lobed particles and the flow in a small size packed bed reactor. Our simulation results are overall consistent with previous experimental work.
|
Page generated in 0.0984 seconds